Retrieval of cloud droplet size from visible and microwave

radiometric measurements during INDOEX: Implication

to aerosols’ indirect radiative effect

Guosheng Liu,1 Hongfei Shao,1 James A. Coakley Jr.,2 Judith A. Curry,3

Julie A. Haggerty,4 and Mark A. Tschudi4

Received 15 October 2001; revised 18 February 2002; accepted 12 April 2002; published 3 January 2003.

[1] The effective radius of water cloud droplets is retrieved using remotely sensed passivemicrowave and visible data collected by aircraft during the Indian Ocean Experiment.The purpose of this study is to assess the aerosols’ effect on cloud microphysical andradiative properties. To study this effect, we investigate the relationships among effectiveradius, liquid water path and number concentration of cloud droplets. The effective radiusretrieval uses imager observations of reflected sunlight at 0.64 mm and liquid water pathderived frommicrowavemeasurements. Results of an error analysis show that retrieval erroris the largest for thin clouds having small visible reflectances and small liquid water path.For this reason, only pixels with visible reflectances greater than 0.2 are used in our dataanalysis, so that the maximum RMS error in effective radius is limited to about 4 mm. Therelation between liquid water path and effective radius is examined for four differentlatitudinal regions. Results show that for the same liquid water path, effective radii aresignificantly smaller in the north than in the south, in correspondence to the north-southgradient of aerosol concentration in this region. In situ aircraft observations reveal largercloud droplet number concentrations in the north than in the south. The north-south gradientof these variables are consistent with the aerosols’ effect on cloud microphysics, that is,higher aerosol concentration increases the number concentration of cloud droplets, which, inturn, reduces droplet sizes given the same liquid water path and cloud thickness. Resultsbased on comparison between data collected from northern and southern hemispheressuggest that the increase in aerosol number concentration alters cloud droplet numbers andsizes while leaving liquid water contents approximately the same. INDEX TERMS: 0305

Atmospheric Composition and Structure: Aerosols and particles (0345, 4801); 0320 Atmospheric Composition

and Structure: Cloud physics and chemistry; 3359 Meteorology and Atmospheric Dynamics: Radiative

processes; 3360Meteorology and Atmospheric Dynamics: Remote sensing; KEYWORDS: cloud effective radius,

liquid water path, aerosol’s indirect effect, Indian Ocean Experiment

Citation: Liu, G., H. Shao, J. A. Coakley Jr., J. A. Curry, J. A. Haggerty, and M. A. Tschudi, Retrieval of cloud droplet size from

visible and microwave radiometric measurements during INDOEX: Implication to aerosols’ indirect radioactive effect, J. Geophys.

Res., 108(D1), 4006, doi:10.1029/2001JD001395, 2003.

1. Introduction

[2] The radiative effect of aerosols arises both from thedirect effect, that is, the direct scattering and absorption ofsunlight by particles [Ramanathan et al., 2001a], and alsofrom the indirect radiative effect, which arises from the

effect of aerosols on cloud microphysical and radiativeproperties of the clouds [e.g., Twomey et al., 1984;Coakley et al., 1987; Radke et al., 1989]. Motivated byincreasing awareness of the importance of both anthro-pogenic and natural aerosols to the Earth’s radiationbudget, the Indian Ocean Experiment (INDOEX) wasdesigned to quantitatively assess the effect of aerosolson radiation through measuring the radiation budget,aerosol concentration, and clouds over the Indian Ocean.The Indian Ocean region was selected for this experimentbecause during the winter monsoon (January–March) theregion provides a nearly ideal natural laboratory forobserving the aerosol radiative forcing. Polluted air flowsoff the Indian and Southeast Asian subcontinents in thenortheasterly flow over the Bay of Bengal and theArabian Sea. The contrast between the high rates ofaerosol emission over the Indian subcontinent and the

JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 108, NO. D1, 4006, doi:10.1029/2001JD001395, 2003

1Department of Meteorology, Florida State University, Tallahassee,Florida, USA.

2College of Oceanic and Atmospheric Sciences, Oregon StateUniversity, Corvallis, Oregon, USA.

3Program in Atmospheric and Oceanic Sciences, University of Colorado,Boulder, Colorado, USA.

4Atmospheric Technology Division, National Center for AtmosphericResearch, Boulder, Colorado, USA.

Copyright 2003 by the American Geophysical Union.0148-0227/03/2001JD001395$09.00

AAC 2 - 1

low rates of emission near the Intertropical ConvergenceZone (ITCZ) south of the subcontinent has been observedto give rise to strong gradients in aerosol concentrations[Rhoads et al., 1997; Jayaraman et al., 1998; Ramana-than et al., 2001b], and is hypothesized to be associatedwith corresponding gradients in cloud microphysical andoptical characteristics.[3] To assess the aerosols’ indirect effect on radiation, the

focus of this study is to retrieve the effective radius of watercloud droplets from simultaneous measurements of micro-wave and visible radiation above the clouds, and examiningwhether aerosol concentration has played a role in deter-mining cloud drop size. The effective radius (re) is definedas the ratio of the third and second moments of the particlesize distribution, that is,

re ¼

R10

r3n rð Þdr

R10

r2n rð Þdr; ð1Þ

where n(r) is the number concentration of particles withradius r. Studies on effective radius retrievals have beenconducted by Nakajima and King [1990], Nakajima et al.[1991], Minnis et al. [1992], Han et al. [1994], Nakajimaet al. [2001] and Szczodrak et al. [2001] using combinationof visible and near-infrared measurements, and by Zuidemaand Hartmann [1995], Greenwald et al. [1995], Lin et al.[1998a, 1998b] and Masunaga et al. [2002] usingcombined visible and microwave measurements. Han etal. [1994] found that the effective radii are generallysmaller for continental water clouds than for marine waterclouds, and smaller in the northern hemisphere than insouthern hemisphere, implying a systematic reduction ofparticle size due to high aerosol concentration. Minnis etal. [1992] reported that the diurnal variation of clouddroplets size observed during the First InternationalSatellite Cloud Climatology Project (ISCCP) RegionalExperiment (FIRE) off the coast of California can beexplained by the diurnally changing wind direction thatbrings polluted continental air in the afternoon and cleanoceanic air in the morning.[4] In a nonprecipitating cumulus cloud, liquid water

path, effective radius and number concentration of clouddroplets are interrelated parameters. For clouds with thesame liquid water path and cloud thickness, the cloud withlower drop number concentration has a larger averageeffective radius. On the other hand, both liquid water pathand cloud droplet size also increase as clouds develop.Consequently, a smaller effective radius near cloud topmay be a result of high number concentration of clouddroplets, but may also be due to the cloud being suppressed.The former reflects the effect of aerosols because they arethe source of cloud condensation nuclei (CCN), while thelatter is apparently related to atmospheric stability and thewater vapor supply. Therefore, to assess the effect ofaerosols on cloud microphysics it becomes necessary toexamine both liquid water path and effective radius, noteffective radius alone. The interrelation among variablesthat are related to effective radius was discussed by Han etal. [1998], who derived the following relation among therelative changes of effective radius (re), liquid water path

(LWP), cloud drop number concentration (Nc), and cloudthickness (h):

�LWP

LWP� 3

�re

reþ�Nc

Nc

þ�h

h: ð2Þ

If we consider a subset of clouds that all have the sameobserved LWP, and we assume the cloud thickness for thissubset being similar, that is, �LWP/LWP = 0 and �h/h � 0,the relation becomes

�Nc

Nc

� �3�re

re: ð3Þ

In other words, under the common value of LWP and theassumption that the cloud thickness for these clouds issimilar, we may examine the change in cloud numberconcentration using the observed change in effective radius.It is noted that the above assumption is equivalent toassuming that mean liquid water content keep constant whilechanges occur. Results based on in situ aircraft microphysicalobservations in this region are consistent with this assump-tion [Heymsfield and McFarquhar, 2001]. This relation canalso be derived for ideal adiabatic clouds while assuming aconstant liquid water path [Brenguier et al., 2000].[5] We use retrievals of LWP and re and in situ aircraft

measurement of cloud drop number concentration to inves-tigate the aerosols’ effect. In a previous study [Liu et al.,2001], liquid water path was retrieved for the INDOEXexperiment using airborne microwave data. In this study, wefocus on the retrieval of effective radius of the cloudsobserved in the INDOEX region, and analyze its relationwith liquid water path.

2. Instrumentation and Data

[6] The Multichannel Radiometer (MCR) and AirborneImaging Microwave Radiometer (AIMR) were deployed onthe NSF/NCAR C-130 aircraft during the 1999 intensiveobservation period of INDOEX to measure upward radia-tions from above the cloud. Eighteen flights were conductedfrom 16 February to 25 March 1999. MCR and AIMR datawere collected, processed, and archived by NCAR’s Atmos-pheric Technology Division. Observations extend from10�S to 10�N mostly between 70� and 75�E except forthe region north of 5�N where there are also data to the westof 70�E. Remotely sensed radiances from the MCR and theAIMR are the primary data source for cloud effective radiusand liquid water path retrievals.[7] The MCR is a seven-channel scanning radiometer

measuring radiances from 0.64 to 10.8 mm [Curran et al.,1981]. We use only data from channel 1 (0.64 mm) in thisstudy; other channels had not been processed for this study.The field of view of MCR is 0.4�, resulting in a footprint onthe ground of about 35 m when the aircraft flies at a nominalaltitude of 5 km. The MCR scans ±45� across nadir with ascan rate of 3.5 times per second. The most recent calibrationof the MCR performed at Los Alamos National Laboratorywas used to relate radiance and voltage count. The radianceat channel 1 is expected to be accurate within ±1.4 W m�2

sr�1 mm�1 [Curran et al., 1981; Tschudi and Laursen, 2001].[8] The AIMR is a dual-frequency, dual-polarization total

power radiometer that measures radiances at 37 and 90 GHz[Collins et al., 1996]. The field of view is 1� for 90 GHz and

AAC 2 - 2 LIU ET AL.: CLOUD EFFECTIVE RADIUS RETRIEVAL DURING INDOEX

2.8� for 37 GHz, resulting in about 90 m and 250 mfootprints, respectively, when the aircraft flies at a nominalaltitude of 5 km. The AIMR scans ±60� across nadir cover-ing a larger area than the MCR although only data where thetwo instruments overlap are analyzed in this study.[9] The retrieval algorithm for liquid water path has been

described by Liu et al. [2001]. It is optimized for AIMRchannels under the INDOEX atmospheric conditions. Thisalgorithm takes into account the better sensitivity of 90 GHzfor thin clouds and the better sensitivity of 37 GHz for thickclouds. Results of analyzing clear-sky scenes show that theretrievals have essentially no bias and a random error of�18 g m�2.[10] Although the MCR and AIMR were onboard the

same aircraft, separate data processing techniques wereused. The data processing techniques for each system usedifferent assumptions to perform calculations of pixel coor-dinates. As a result, the coordinates of data points do notmatch well with each other between the two instruments.This mismatch can be easily noticed by examining theimages of raw data. The following procedure is introducedin order to obtain the best collocation. First, all observationsare divided into small ‘‘scenes’’ with flight duration nomore than 5 min. Within each scene, correlation coefficientis calculated between liquid water path and 0.64 mmradiance, first based on the coordinates provided by theraw data sets, and then shifting the MCR coordinates incross-scan and along-track directions by 20 m each time.This procedure continues until a clear maximum correlationis found, and the best collocation match is determined.[11] Since the pixel size of AIMR is several times larger

than that of MCR, a weighted average is used to combineseveral MCR pixels to match an AIMR pixel. For a givenAIMR pixel, we first select the MCR pixels that have adistance between their center and the center of the AIMRpixel less than half of the pixel size at AIMR 37 GHz. Theweight assigned to a selected MCR pixel decreases with thedistance of its center to the center of the AIMR pixelfollowing the function exp(�d2/2s2), where d is the dis-tance and s is a constant and can be determined by lettingthe weight reduce to half when d increases from 0 to half ofthe pixel size at AIMR 37 GHz. The weights are thennormalized so that the sum of the weights of all selectedMCR pixels equals unity. This analysis procedure effec-tively creates visible reflectances having the same spatialresolution as the AIMR brightness temperatures and beingcollocated with the AIMR field of view.[12] For the retrievals of liquid water path and effective

radius, only the data collected when the aircraft was flyinglevel over cloud top are usable. As mentioned above, dataare divided into scenes of less than 5 min each for datacollocation purpose. As will be discussed below, sceneswere restricted to contain low-level cumulus clouds havinghorizontal scales >2 km. Of all the available data, thirty-fivecloudy scenes were found that satisfied these conditions.The locations of the scenes are shown in Figure 1.

3. Cloud Effective Radius Retrieval

3.1. Description of the Retrieval Algorithm

[13] The effective radius retrieval is based on a lookuptable generated in advance by a set of atmospheric radiative

transfer simulations. The inputs for the lookup table areliquid water path retrieved from microwave data and visiblereflectances at 0.64 mm defined by

R ¼ pIobsm0F0

; ð4Þ

where Iobs is the observed radiance, m0 is the cosine ofsolar zenith angle and F0 is the incident solar radiativeflux at 0.64 mm. Apart from cloud microphysical proper-ties, the reflected solar radiance also depends on solarzenith and azimuth angles, sensor viewing zenith andazimuth angles, and cloud altitude. In practice, many suchlookup tables are produced, varying solar and sensorviewing angles by 5� and cloud top altitude by 0.5 km.Cloud base is assumed at 0.6 km, which is the typicalconvective condensation level in the region. Cloud topheight is estimated from cloud top temperature measuredby the nadir looking narrow band infrared radiometeronboard the C-130.[14] The radiative transfer model used in the retrieval is

the SBDART (Santa Barbara DISORT Atmospheric Radia-tive Transfer) model developed by Ricchiazzi et al. [1998].This model is a combination of a discrete ordinate radiativetransfer module [Stamnes et al., 1988], low-resolutionatmospheric transmission models, and Mie scattering for-mula for light scattering by water droplets and ice crystals.A modified gamma size distribution is used for clouddroplets:

n rð Þ ¼ cr

re

� �p�1

exp � pþ 2

rer

� �; ð5Þ

where n(r) is the number concentration of cloud dropletswith radius r, c is a normalization constant, p is a

Figure 1. Locations of the thirty-five cloudy scenesanalyzed in this study.

LIU ET AL.: CLOUD EFFECTIVE RADIUS RETRIEVAL DURING INDOEX AAC 2 - 3

dimensionless parameter that controls the width of thedistribution (p = 7 is used in the SBDART model). Detaileddescription of the model design and validation can be foundin Ricchiazzi et al. [1998]. In the radiative calculations, astandard tropical atmosphere [McClatchey et al., 1972] with100% relative humidity in the cloud layer and a Lambertianocean surface [Tanre et al., 1990] are used. The oceansurface albedo at 0.64 mm is �0.045. A standard maritimeaerosol profile is used for all model runs. The aerosoloptical depth at 0.64 mm for this profile is about 0.25. Whilethe surface albedo and aerosol optical depth are taken to beconstant throughout the INDOEX domain, the minimumcloud reflectance, 0.2, and optical depth, �3, used in thisstudy make the effects of varying surface albedo andaerosol optical depths negligible in this study. Because ofthe plane-parallel assumption, the three-dimensional effectsare not included in the radiative transfer modeling, whichconstitutes a source of uncertainty in the effective radiusretrieval, particularly for observations with large solarzenith angles.[15] The basic concept underlying the determination of

effective radius is illustrated in Figure 2. The figure showsthe relationship between cloud liquid water path and 0.64mm reflectances calculated with SBDART. Results areshown for effective radii from 2 to 32 mm and the reflec-tances were calculated at aircraft altitude. These relationswere calculated based on nadir observations at 0653 to0658Z on 27 February 1999, near 9�N and 66.5�E. OnceLWP is determined from AIMR brightness temperatures, thecorresponding 0.64 mm reflectance is used to determine theeffective radius. This diagram shown in Figure 2 serves as alookup table. Retrieved re are obtained through interpolationof the reflectances for a given LWP. Also shown in thisfigure are all available data points at pixel level (small opencircles), data points for which retrievals were performed

after screening (small solid circles), and the average of allretrievable points over each 0.05 interval in the 0.64 mmreflectance (large solid circles). On average, effective radiivalues for this case range from 6 to 12 mm while liquidwater path values vary from 0 to 200 g m�2. The reflec-tances shown in Figure 2 suggest that the MCR saturatedwhen the reflected sunlight was large. Consequently,detailed analyses are restricted to clouds with liquid waterpaths less than about 200 g m�2.

3.2. Error Analysis

[16] Since cloud drop effective radii are retrieved by thecombination of microwave LWP and visible reflectances,we first examine the error characteristics of these two inputvariables. Due to the nonexistence of coincident validationdata, the error characterization is done by analyzing theobserved data for cloud-free regions. The frequency distri-bution of retrieved LWP for cloud-free regions is shown inFigure 3a, which indicates an almost zero mean and an 18 gm�2 standard deviation. That is, the systematic and randomerrors in the LWP retrieval are approximately 0 and 18 gm�2, respectively. Similarly, we analyzed the differencebetween observed and radiative transfer model calculated

Figure 2. Relationship between liquid water path andreflectance (0.64 mm) calculated by a radiative transfermodel for cloud effective radii from 2 to 32 mm (curves).The calculation is based on nadir observations at 0653 to0658Z on 27 February 1999. Small open circles are allobservational data at pixel level, small solid dots are thosepixels that retrievals are applied, and large solid circles arethe average of retrieved data at each 0.05 reflectance bin.

Figure 3. Error characteristics of (a) liquid water path and(b) reflectance derived from observations at clear-sky areas.

AAC 2 - 4 LIU ET AL.: CLOUD EFFECTIVE RADIUS RETRIEVAL DURING INDOEX

reflectances for clear-sky regions (Figure 3b). The meanreflectance error is negligible and the standard deviation is0.016, which is a measure of the random error. This errorincludes instrument noise and the uncertainties of variablesused in the radiative transfer model, such as surface reflec-tion, aerosol optical depth, etc.[17] The random error in effective radii caused by the

uncertainties in the input variables is then estimated by aMonte Carlo procedure. We randomly add noise to theinput values within the range of plus and minus onestandard deviation, that is, ±18 g m�2 for LWP and±0.016 for reflectances, and compare the ‘‘retrieved’’ tothe true (input) re. A similar procedure was used by Liuand Curry [2000] in analyzing random error in ice waterpath retrievals. Figure 4 shows the root mean square (rms)error after randomly perturbing the input variables for 106

times. The biggest error occurs when both LWP andreflectances are small (with the maximum error greaterthan 10 mm), and the RMS error decreases to less than 4 mmas LWP larger than 50 g m�2 or reflectances larger than0.2. Clearly, the retrieval method is more suitable forrelatively thick nonprecipitating clouds. Considering thelarge error associated with thin clouds, we use only pixelsthat have 0.64 mm reflectances greater than 0.2 regardlessof viewing geometry in all the following discussions if nototherwise indicated, which limits the maximum RMS errorin re to about 4 mm.[18] Another important error that was not included in the

above error analysis arises from the fact that cloudsobserved during INDOEX are boundary layer cumulusand have small horizontal scales, so that the plane-parallelassumption in our radiative transfer calculations is notstrictly met. This error is very difficult to assess althoughit should decrease as the cloud’s horizontal scale increases.For this reason, in our analysis we excluded the scenesthat only have small clouds with horizontal scale less than2 km.[19] In Figure 5, we show one example of the retrievals.

The images shown are liquid water path derived from

microwave data (left), reflected visible radiance at 0.64mm (middle) and retrieved effective radius (right) for 0653to 0658Z on 27 February 1999, near 9�N and 66.5�E. Thescene covers an area of approximately 12 km by 50 km andall available pixels including those with reflectance less than0.2 are shown. The convective cells correspond well amongthe three images although the microwave LWP misses somesmall cells. The retrieved effective radius ranges from nearzero to �30 mm, with the largest value near the center ofconvective cells. Although there is no in situ measurementavailable to validate the re retrievals, the reasonable patternshown here suggests that the algorithm can produce at leastqualitatively correct results.

4. Results

4.1. A Clean Case and a Polluted Case

[20] We start the analysis by examining two casesobserved in a clean and a polluted region, respectively.The first case was observed on 4 March 1999 near 9.3�S,71.0�E (RF08), and the second case was observed on 27February 1999 near 9�N, 66.5�E (RF06). The sizes of thecloud cells are comparable between the two cases. Theaveraged re for pixels with reflectance greater than 0.2 isabout 15 and 10 mm for RF08 and RF06 case, respectively.[21] Figure 6 shows the LWP-re relationship by fre-

quency distribution for the two cases. The frequency iscalculated by the pixel number within each 30 g m�2LWP

Figure 4. Root mean square error in effective radiusretrievals estimated using a Monte Carlo procedure. Unit:mm. It is noted that the largest error occurs when both liquidwater paths and visible reflectances are small.

Figure 5. Images of liquid water path (left), reflectedradiance at 0.64 mm (middle) and effective radius (right) forthe cloudy scene of 0653 to 0658Z on 27 February 1999.The scene covers an area of approximately 12 km by 50 km.

LIU ET AL.: CLOUD EFFECTIVE RADIUS RETRIEVAL DURING INDOEX AAC 2 - 5

and 3 mm re bin, which is then normalized by the totalpixel number and multiplied by 1000. The general trend ofthe LWP-re relation is that re increases with LWP, whichmay be predicted by an adiabatic cloud model [e.g.,Brenguier et al., 2000]. In the figure, the solid dots showthe modal effective radii that are determined by themaximum frequency at fixed liquid water path. Despitethe scatter in the frequency distribution, we can still findthe general trend by following the modal effective radiusas LWP increases. From this trend, it is seen that reincreases more sharply in the RF08 case than in theRF06 case. For example, the modal re increases from 3to 20 mm as LWP changes from 0 to150 g m�2 in theRF08 case, while it only increases from 3 to 13 mm for thesame LWP variation in the RF06 case.[22] The difference in cloud microphysics between the

two cases is further shown by the re frequency distribu-tions at fixed liquid water path as shown in Figure 7. Thefrequency is normalized by the total pixel number at thegiven liquid water path bin (25 g m�2) so that the inte-grated frequency for all effective radii is 100. Except forLWP = 25 g m�2, the cloud in the clean case (RF08)shows systematically larger effective radii than the cloudin the polluted case (RF06). The difference in modaleffective radii between the two cases exceeds 5 mm forLWP of 100 and 150 g m�2. In other words, for the samevalue of LWP, clouds in the RF06 case are more likely tocontain smaller water droplets than clouds in the RF08case. As mentioned earlier, the northern region is pollutedand thus contains more aerosols. The smaller droplet sizemay be interpreted as a result of the high numberconcentration of CCN.[23] Limited particle number concentration data were

collected by the Forward Scattering Spectrometer Probe100 (FSSP-100) and 300 (FSSP-300) [Dye and Baumgard-ner, 1984]. The FSSP-100 measures cloud particles rangingfrom 2 to 47 mm while the FSSP-300 measures particlesranging from 0.3 to 20 mm, both with a precision of 16% fornumber concentration. The number concentration obtainedfrom the FSSP-100 and FSSP-300 is plotted in Figure 8,corresponding to the two cases. Note that the FSSP data arenot necessarily sampled from the same clouds as thoseshown for Figure 6 and 7. Our assumption is that thenumber concentrations of any clouds in the same area aresimilar except near the cloud edges where the cloud water isdiluted by mixing with surrounding drier air. In the RF08case, the aircraft sampling was conducted near the cloudbase of 550 m altitude while the sampling was done throughthe layer between 700 and 1300 m in the RF06 case. Fornondrizzling boundary layer cumulus clouds, cloud dropconcentration does not vary much throughout the cloud’svertical extent. In other words, the value sampled near cloudbase in the RF08 case also reflects the cloud drop concen-tration throughout the entire cloud. It is also reasonable toconsider that the maximum values in the number concen-tration profiles reflect the droplet concentration where thecloud water has not been diluted by dry air. Therefore, itappears that there is a significant difference in cloud dropletnumber concentrations between clouds in the two cases. Forexample, the maximum value of FSSP-100 counts is �100cm�3 in the RF08 case while it is �400 cm�3 in the RF06case. Recalling that the southern region (RF08) is far from

the pollution source of Indian subcontinent, the differencebetween the two cases is a strong indication of the aerosols’effect on cloud microphysics.

4.2. Relationship Between LWP and reObserved During INDOEX

[24] Using data from the 35 scenes shown in Figure 1, theaveraged LWP and re values are calculated for each 5�-latitude region. Note that there is only one scene in the 0–5�N region and two scenes in the 0–5�S region. Thelatitudinal distribution and scatterplot (inset) of LWP and reare shown in Figure 9. Only pixels with reflectance greaterthan 0.2 are included in the averaging. The clouds

Figure 6. Frequency distribution of pixel occurrence inthe two-dimensional diagram of liquid water path andeffective radius for (a) the clean case on 4 March 1999(RF08) near 9.3�S, 71.0�E and (b) the polluted case on 27February 1999 (RF06) near 9.0�N, 66.5�E. Solid dots showthe modal effective radii at fixed liquid water path.

AAC 2 - 6 LIU ET AL.: CLOUD EFFECTIVE RADIUS RETRIEVAL DURING INDOEX

observed in the southern part of the domain both havegreater LWP and larger re than those observed in the north.The scatterplot shows the positive correlation betweenLWP and re.[25] The frequency distributions of the LWP-re relation

for the 4 latitudinal regions are shown in Figure 10, whichare calculated using the same procedure as that for the casestudies previously discussed. The solid dots, showing themodal effective radii at fixed LWP, illustrate the generaltrend of the LWP-re relation. For LWP value of 100 g m�2,the most probable re is around 10 mm in the northern regionswhile larger than 15 mm in the southernmost region.[26] To further illustrate the latitudinal difference, we

calculated the frequency distribution of effective radii atfixed LWPs, and the modal, median and mean effective radiias a function of LWP for the northernmost (5–10�N) andthe southernmost (5–10�S) regions. These two regionscontain more data points than the other two regions. Theresults are shown in Figures 11 and 12. There is a cleardifference in frequency distributions of effective radiusbetween the two regions (Figure 11). That is, given a fixedvalue of LWP, clouds in the south tend to have larger

effective radii than clouds in the north. This difference isalso shown in modal, median, and mean radii plots (Figure12), particularly for clouds in the large LWP range. Thebiggest difference is shown in the modal radius while themean radius shows smaller differences. This result is anindication that high aerosol concentration in the north hasreduced the cloud droplet size by increasing numberconcentration of cloud droplets. For small LWP (e.g.,

Figure 8. Particle number concentrations measured byFSSP-100 and FSSP-300 probes for (a–b) the clean caseand (c–d) the polluted case.

Figure 9. Latitudinal distribution of liquid water path andeffective radius averaged in each 5� latitudinal region. Inset:Scatterplot of the averaged liquid water path and effectiveradius.

Figure 7. Comparison of frequency distributions ofeffective radius at liquid water path of (a) 25 g m�2, (b)50 g m�2, (c) 100 g m�2, and (d) 150 g m�2 between theclean case (RF08) and the polluted case (RF06).

LIU ET AL.: CLOUD EFFECTIVE RADIUS RETRIEVAL DURING INDOEX AAC 2 - 7

LWP = 25 g m�2), the difference in the effective radiusbetween the two regions is less clear. Although the smalldifference may be a true reflection of the real clouds, thelarge uncertainty in the effective radius retrieval in the smallLWP range (c.f. Figure 4) makes it difficult to draw areliable conclusion.

4.3. North-South Gradient of Cloud DropletsConcentration From In Situ Data

[27] As mentioned in the introduction, the INDOEXregion is characterized by strong north-south gradient ofanthropogenic aerosols as a result of the polluted airmassover the Indian subcontinent and the clean airmass overtropical oceans. The difference in LWP-re relations describedearlier suggests that clouds in the INDOEX domain haveresponded to this aerosol gradient, that is, smaller cloudradii to the north and larger cloud radii to the south, giventhe same values of LWP. To further elaborate this point, weanalyze the in situ data of cloud drop concentrationsmeasured by FSSP-100 on C-130 aircraft. Within every 1�(latitude) by 1� (longitude) box, we average the highest 10values registered on FSSP-100s one-second number con-centration data during any of the 18 flights. The highestvalues are used on the assumption that they arise frommeasurements well within a cloud where cloud water hasnot been diluted by surrounding drier air. Figure 13 showsthe distribution of the averaged values of the FSSP-100cloud drop number concentration. The highest concentra-tion was observed near the northeast corner of the regionwith a value of � 500 cm�3. The concentration decreasestoward the south to � 100 cm�3 at the southernmost extentof the region. Again, this diagram suggests that the aerosolconcentration is affecting the number concentration of cloud

droplets in the INDOEX domain, which is consistent withour results based on LWP and re retrievals.[28] Using the in situ aircraft cloud drop number concen-

tration and retrievals of liquid water path and effectiveradius, we estimated the magnitude of terms shown in (2)by taking the differences of LWP, re and Nc betweennorthern and southern hemispheres. The relative differencein cloud number concentration estimated from data inFigure 13 is �Nc /Nc � 0.6. From data shown in Figure 9,the relative differences in effective radius and liquid waterpath were estimated as 3�re /re � �0.5 and �LWP/LWP ��0.1. Consequently, it appears that the increase in clouddroplet number concentration from southern to northernhemispheres is roughly balanced by the decrease in effectiveradius of cloud droplets while leaving the liquid watercontents approximately the same. Similar findings werereported by Heymsfield and McFarquar [2001] based solelyon in situ aircraft microphysical observations. Additionally,the �10% difference in liquid water path together with the�10% imbalance between �Nc /Nc and 3�re /re suggeststhat the average cloud thickness in the northern hemisphereis �20% smaller than that in the southern hemisphere. This

Figure 10. Frequency distribution of pixel occurrence inthe two-dimensional diagram of liquid water path andeffective radius for the following four latitudinal regions. (a)5–10�S, (b) 0–5�S, (c) 0–5�N, and (d) 5–10�N. Solid dotsshow the modal effective radii at fixed liquid water path.

Figure 11. Comparison of frequency distributions ofeffective radius at liquid water path of (a) 25 g m�2, (b)50 g m�2, (c) 100 g m�2, and (d) 150 g m�2 between 5–10�S and 5–10�N regions.

AAC 2 - 8 LIU ET AL.: CLOUD EFFECTIVE RADIUS RETRIEVAL DURING INDOEX

difference may be a reflection of the differences inatmospheric thermodynamical structures between the twohemispheres and/or due to aerosols’ cloud thinning effect assuggested by Ackerman et al. [2000].

5. Conclusions

[29] The effective radius of water cloud droplets wasretrieved using remotely sensed microwave and visible datacollected by airborne radiometers during the INDOEXexperiment. The INDOEX domain is characterized by astrong north-south gradient of aerosol concentration due tothe contrast of the polluted airmass from Indian subcon-tinent and the clean airmass in the ITCZ. This study aims toassess whether the aerosol concentration gradient gives rise

to the difference in effective radii of cloud droplets betweenthe northern and southern regions.[30] The effective radius is retrieved from the reflectance

at 0.64 mm and liquid water path derived from microwaveobservations. Error analysis using a Monte Carlo procedureshows that the largest error occurs for thin clouds havingsmall 0.64 mm reflectances and small liquid water paths. Forthis reason, only pixels with a reflectance larger than 0.2 areused in the data analysis, which limits the maximum RMSerror in effective radius to about 4 mm. To minimize theerror caused by horizontal inhomogeneity of cloud field, weexcluded the cloudy scenes that only have clouds less than 2km in horizontal scale.[31] Thirty-five cloudy scenes were selected for this

study, which were located approximately along 70�E from10�S to 10�N. The averaged liquid water path and effectiveradius both show north-south gradients with higher valuesin the southern regions. We first investigated two cases, onein clean air and the other in polluted air. For the same valueof liquid water path, effective radii are smaller in thepolluted case than in the clean case. Investigation for fourlatitudinal regions shows that for the same liquid water path,effective radii are significantly smaller in the northernregions than in the southern regions, corresponding to theaerosols’ north-south gradient. In situ aircraft measurementby FSSP-100 also shows the north-south gradient of clouddroplet number concentration. The north-south differencesin liquid water path, effective radius, and cloud numberconcentration are consistent with the hypothesis of theaerosols’ effect on cloud microphysics, that is, higheraerosol concentration increases the number concentrationof cloud droplets, which in turn reduces droplet sizes giventhe same liquid water path and cloud thickness. Further-more, a rough estimate of the differences in effective radiiand cloud number concentrations between the two hemi-

Figure 12. Relations between liquid water path and (a)modal radius, (b) median radius, and (c) mean radius for thesouthern (5–10�S) and northern (5–10�N) regions.

Figure 13. Distribution of particle number concentrationmeasured by FSSP-100. The number concentration shownhere is the average of the highest 10 values in each 1� 1�grid observed during INDOEX experiment.

LIU ET AL.: CLOUD EFFECTIVE RADIUS RETRIEVAL DURING INDOEX AAC 2 - 9

spheres suggests that the aerosol number concentrationalters cloud droplet numbers and sizes while leaving liquidwater contents approximately the same. Due to the limita-tion of aircraft measurements, further research efforts usingsatellite observations are needed to extend the investigationto global and long timescales. Additionally, collectingground truth data in future field experiments will be veryvaluable to validate liquid water path and effective radiusretrievals.

[32] Acknowledgments. We thank scientists and engineers at NCAR,especially Craig Walther and Krista Laursen, for collecting the AIMR andMCR data during INDOEX and processing them after the experiment.Comments and suggestions from Qingyuan Han at the University ofAlabama in Huntsville and two anonymous reviewers were very helpful.This research has been supported by NSF Grant ATM-0002860 and NASAGrant NAG5-8647. J. Curry’s participation was supported by the NASAFIRE project. J. Coakley’s participation was supported by NSF through theCenter for Clouds, Chemistry, and Climate (C4), a national science andtechnology center at the Scripps Institution of Oceanography and by NSFGrant ATM-9612886.

ReferencesAckerman, A. S., O. B. Toon, D. E. Stevens, A. J. Heymsfield, V. Rama-nathan, and E. J. Welton, Reduction of tropical cloudiness by soot,Science, 288, 1042–1047, 2000.

Brenguier, J.-L., H. Pawlowska, L. Schuller, R. Preusker, J. Fischer, and Y.Fouquart, Radiative properties of boundary layer clouds: Droplet effec-tive radius versus number concentration, J. Atmos. Sci., 57, 803–821,2000.

Coakley, J. A., Jr., R. L. Bernstein, and P. A. Durkee, et al., Effect of ship-track effluents on cloud reflectivity, Science, 237, 1020–1022, 1987.

Collins, M. J., F. G. R. Warren, and J. L. Paul, Airborne imaging microwaveradiometer - Part I: Radiometric analysis, IEEE Trans. Geosci. RemoteSens., 34, 643–655, 1996.

Curran, R. J., H. L. Kyle, L. R. Blaine, J. Smith, and T. D. Clem, Multi-channel scanning radiometer for remote sensing cloud physical para-meters, Rev. Sci. Instrum., 52, 1546–1555, 1981.

Dye, J. E., and D. Baumgardner, Evaluation of the forward scatteringspectrometer probe Part I: Electronic and optical studies, J. Atmos. Ocea-nic. Technol., 1, 330–344, 1984.

Greenwald, T. J., G. L. Stephens, S. A. Christopher, and T. H. Vonder Haar,Observations of the global characteristics and regional radiative effects ofmarine cloud liquid water, J. Clim., 8, 2928–2946, 1995.

Han, Q., W. B. Rossow, and A. A. Lacis, Near-global survey of effectivedroplet radii in liquid water cloud using ISCCP data, J. Clim., 7, 465–497, 1994.

Han, Q., W. B. Rossow, J. Chou, and R. M. Welch, Global survey of therelationships of cloud albedo and liquid water path with droplet size usingISCCP, J. Clim., 11, 1516–1528, 1998.

Heymsfield, A. J., and G. M. McFarquhar, Microphysics of INDOEX cleanand polluted trade cumulus clouds, J. Geophys. Res., 106, 28,653–28,674, 2001.

Jayaraman, A., D. Lubin, S. Ramachandran, V. Ramanathan, E. Wood-bridge, W. Collins, and K. S. Zalpuri, Direct observations of aerosolradiative forcing over the tropical Indian Ocean during the Jan.-Feb.-1996 Pre-INDOEX cruise, J. Geophys. Res., 103, 13,827–13,836, 1998.

Lin, B., B. Wielicki, P. Minnis, and W. B. Rossow, Estimation of watercloud properties from satellite microwave, infrared and visible measure-ments in oceanic environments, 1, Microwave brightness temperaturesimulations, J. Geophys. Res, 103, 3873–3886, 1998a.

Lin, B., P. Minnis, B. Wielicki, D. R. Doelling, R. Palikonda, D. F. Young,and T. Uttal, Estimation of water cloud properties from satellite micro-wave, infrared and visible measurements in oceanic environments, 2,Results, J. Geophys. Res., 103, 3887–3905, 1998b.

Liu, G., and J. A. Curry, Determination of ice water path and mass medianparticle size using multichannel microwave measurements, J. Appl. Me-teorol., 39, 1318–1329, 2000.

Liu, G, J. A. Curry, J. A. Haggerty, and Y. Fu, Retrieval and characteriza-tion of cloud liquid water path using airborne passive microwave dataduring INDOEX, J. Geophys. Res., 106, 28,719–28,730, 2001.

Masunaga, H., T. Y. Nakajima, T. Nakajima, M. Kachi, R. Oki, and S.Kuroda, Physical properties of maritime low clouds as retrieved by com-bined use of Tropical Rainfall Measurement Mission Microwave Imagerand Visible/Infrared Scanner: Algorithm, J. Geophys. Res., 107, 4083,doi:10.1029/2001JD000743, 2002.

McClatchey, R. A., R. W. Fenn, J. E. A. Selby, F. E. Volz, and J. S. Garing,Optical properties of the atmosphere, 3rd ed., AFCRL Environ. Res.Pa-pers No. 411, 108 pp., 1972.

Minnis, P., P. W. Heck, D. F. Young, C. W. Fairall, and J. B. Snider,Stratocumulus cloud properties derived from simultaneous satellite andisland-based instrumentation during FIRE, J. Appl. Meteorol., 31, 317–339, 1992.

Nakajima, T., and M. D. King, Determination of the optical thickness andeffective particle radius of clouds from reflected solar radiation measure-ments, Part I, Theory, J. Atmos. Sci., 47, 1878–1893, 1990.

Nakajima, T., M. D. King, and J. D. Spinhirne, Determination of the opticalthickness and effective particle radius of clouds from reflected solarradiation measurements, Part II, Marine stratocumulus observations,J. Atmos. Sci., 48, 728–750, 1991.

Nakajima, T., A. Higurashi, K. Kawamoto, and J. E. Penner, A possiblecorrelation between satellite-derived cloud and aerosol microphysicalparameters, Geophys. Res. Lett., 78, 1171–1174, 2001.

Radke, L. F., J. A. Coakley Jr., and M. D. King, Direct and remote sensingobservations of the effects of ships on clouds, Science, 246, 1146–1149,1989.

Ramanathan, V., P. J. Crutzen, J. T. Kiehl, and D. Rosenfeld, Aerosols,climate, and the hydrological cycle, Science, 294, 2119–2124, 2001a.

Ramanathan, V., et al., Indian Ocean Experiment: An integrated assessmentof climate forcing and effects of the great Indo-Asian haze, J. Geophys.Res., 106, 28,371–28,398, 2001b.

Rhoads, K., P. Keley, R. Dickerson, T. Carsey, M. Farmer, D. Savoi, and J.Propero, The composition of the troposphere over the Indian Ocean dur-ing the monsoonal transition, J. Geophys. Res., 102, 18,981–18,995,1997.

Ricchiazzi, P., S. Yang, C. Gautier, and D. Sowle, SBDART: A research andteaching software tool for plane-parallel radiative transfer in the Earth’satmosphere, Bull. Am. Meteorol. Soc., 79, 2101–2114, 1998.

Stamnes, K., S. Tsay, W. Wiscombe, and K. Jayaweera, Numericallystable algorithm for discrete-ordinate-method radiative transfer in multi-ple scattering and emitting layered media, Appl. Opt., 27, 2502–2509,1988.

Szczodrak, M., P. H. Austin, and P. B. Krummel, Variability of opticaldepth and effective radius in marine stratocumulus clouds, J. Atmos.Sci., 58, 2912–2926, 2001.

Tanre, D., C. Deroo, P. Duhaut, M. Herman, J. J. Morcrette, J. Perbos, andP. Y. Deschamps, Description of a computer code to simulate the satellitesignal in the solar spectrum: The 5S code, Int. J. Remote Sens., 11, 659–668, 1990.

Tschudi, M. A., and K. K. Laursen, Airborne spectral reflectance observa-tions at SHEBA from NCAR’s Multichannel Cloud Radiometer (MCR),Paper presented at the Sixth Conference on Polar Met and Oceanography,Am. Meteorol. Soc., San Diego, Calif., 147–150, 2001.

Twomey, S., M. Piepgrass, and T. L. Wolfe, An assessment of the impact ofpollution on global cloud albedo, Tellus, 36B, 356–366, 1984.

Zuidema, P., and D. L. Hartmann, Satellite determination of stratus cloudmicrophysical properties, J. Clim., 8, 1638–1657, 1995.

�����������������������J. A. Coakley Jr., College of Oceanic and Atmospheric Sciences, Oregon

State University, Corvallis, OR 97331, USA. (coakley@oce.orst.edu)J. A. Curry, Program in Atmospheric and Oceanic Sciences, University of

Colorado, Campus Box 429, Boulder, CO 80309, USA. (curryja@cloud.colorado.edu)G. Liu and H. Shao, Department of Meteorology, Florida State

University, 404 Love Building, Tallahassee, FL 30306-4520, USA.(liug@met.fsu.edu; soar@met.fsu.edu)J. A. Haggerty and M. A. Tschudi, Atmospheric Technology Division,

National Center for Atmospheric Research, Boulder, CO 80307, USA.( julie@raf.atd.ucar.edu; tschudi@ucar.edu)

AAC 2 - 10 LIU ET AL.: CLOUD EFFECTIVE RADIUS RETRIEVAL DURING INDOEX